Opus 5.5 vs GPT-6 Sol: 3D Pelican riding bike test in Blender

Over the last few months I've seen the "pelican riding a bicycle" SVG test a bunch of times in my feed and I started wanted to know if models are so good at this point that they can do the same thing, but in 3D, cause every big model can draw a decent pelican in SVG at this point both models got the exact same long art brief: australian pelican on a teal touring bike, boardwalk at sunset, a mackerel in its beak with water dripping off it, wing tips on the handlebars, sun shining through the pouch. they had to build the whole thing from an empty scene with python scripts, no downloaded assets or anything gpt 6 sol finished faster than opus 5.5, but at the same time the result left a lot to be desired. from a distance it looks sorta fine, but if you look closer it really starts to fall apart, the ik rig on the legs is messed up(they literally bend outwards and the feathers rotate with the leg there 😭)and the overall environment art design was not that good, felt empty and not that lively opus 5.5 landed the atmosphere of the shot and the minor details, it also didn't completely ruin the legs(even though there is a small mistake with the base of the legs sticking out of the body when rotating, but that wasn't something that I noticed at first), the color scheme was very enjoyable and small splash effects from the fish in the mouth looked nice. I overall liked opus's result much more than sol's, despite seeing a lot of "leaked" footage from sol which looked stunning, didn't live up to the hype after all :( price wise: Model Cost How it ran Opus 5.5 $48.11 Claude via API key GPT-6 Sol $34.55 Codex sub (this is the API equivalent) so I guess models are finally good enough to do the pelican test in 3D, though the difference can still be pretty big though, as seen in the video setup: claude via api key and codex via sub connected to blender with blendermcp inside atomic.chat